I’m doing some metaprogramming for Interpolations.jl, and I’m getting stuck
at an error that I think is due to scoping issues in my nested quotes, and
I’ve stared at this for so long now I’m not getting anywhere.
I can do the following
matching(d) = quote
println("Matching: ", $d)
end
nonmatching(d) = quote
println("Not matching: ", $d)
end
@ngenerate N T function example{T,N}(A::Array{T,N}, xs::NTuple{N,Real}...)
@nexprs N dim->begin
@nexprs N d->begin
if d==dim
eval(matching(d))
else
eval(nonmatching(d))
end
end
end
end
example(rand(2,2), 1.5, 0.7)
and get the expected output:
Matching: 1
Not matching: 2
Not matching: 1
Matching: 2
Examining with macroexpand, I see that the eval calls are still there, but
for performance reasons I want to avoid them, so I try to rewrite it to use
eval(ngenerate(...)) instead of @ngenerate ...:
#matching and nonmatching defined as before
ex = ngenerate(
:N,
:T,
:(example{T,N}(A::Array{T,N}, xs::NTuple{N,Real}...)),
N->quote
@nexprs $N dim->begin
@nexprs $N d->begin
if d==dim
$(matching(d))
else
$(notmatching(d))
end
end
end
end
)
eval(ex)
example(rand(2,2), 1.5, 0.7)
But this doesn’t work: I get ERROR: d not defined on $(matching(d)) (and,
if I comment that one out, $(nonmatching(d))). I’ve tried $(matching($d))
(ERROR:
error compiling anonymous: syntax: prefix $ in non-quoted expression) as
well as $(matching(esc(d))) and $(esc(matching(d))) (both ERROR: d not
defined), and I’m at a loss for what to try (these errors are all thrown at
the stage of *defining* the quoted expression, so macroexpand hasn’t helped
me here either, as I don’t get any expression to call it on…).
Eventually, this will be used to do different things for different array
dimensions like here, but with both matching and nonmatching also
dispatching on types of some additional arguments that I left out for
simplicity, and choosing expressions by means of multiple dispatch this way
is really at the core of Interpolations.jl, so I can’t move the contents of
matching and nonmatching out of their respective functions.
Thanks in advance for any ideas that get me forward,
// Tomas
